One bounded HTTP client for every server-side fetch; uploads refused while streaming

Every outbound fetch - Gitea content at boot and on reload, the raw
asset proxy, the repo-info endpoint, the content-declared url resource
- built its own reqwest client with no timeout and read whole bodies
with .text(). `crate::http` is now the one client (5 s connect, 20 s
per request, 5 redirects) with capped body readers: 2 MB for a url
resource, 8 MB for a Gitea reply, 32 MB for a proxied asset; one byte
over is an error, never a truncated document.

The upload handler read a whole field into memory before comparing it
to the 20 MB cap, and axum's 2 MB default body limit had made that cap
unreachable anyway. It now reads chunk by chunk and refuses the moment
the cap is crossed; the /upload route raises the body limit to match.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01L4jrCgLiKKHAEFuZUJjckH

src/http.rs

@@ -0,0 +1,64 @@
+//! The one outbound HTTP client, bounded. Every server-side fetch -
+//! Gitea content at boot and on reload, the raw-asset proxy, the
+//! repo-info endpoint, and the content-declared `url` resource - goes
+//! through here, so a slow or oversized upstream can cost a request at
+//! most `TIMEOUT` and `cap` bytes, never a thread or the heap. Before
+//! this module each call site built its own client with no timeout and
+//! read whole bodies with `.text()`.
+#![cfg(feature = "ssr")]
+
+use std::sync::OnceLock;
+use std::time::Duration;
+
+use openidconnect::reqwest;
+
+/// Whole-request budget: connect, headers, and body.
+pub const TIMEOUT: Duration = Duration::from_secs(20);
+const CONNECT_TIMEOUT: Duration = Duration::from_secs(5);
+
+/// A content-declared URL resource: JSON for one card.
+pub const URL_RESOURCE_CAP: usize = 2 * 1024 * 1024;
+/// A Gitea API reply or one content file (a tree listing, a YAML page).
+pub const GITEA_CAP: usize = 8 * 1024 * 1024;
+/// A content-shipped asset proxied to the browser (an image, a font).
+pub const ASSET_CAP: usize = 32 * 1024 * 1024;
+
+pub fn client() -> &'static reqwest::Client {
+    static CLIENT: OnceLock<reqwest::Client> = OnceLock::new();
+    CLIENT.get_or_init(|| {
+        reqwest::Client::builder()
+            .connect_timeout(CONNECT_TIMEOUT)
+            .timeout(TIMEOUT)
+            .redirect(reqwest::redirect::Policy::limited(5))
+            .user_agent(concat!("portal/", env!("CARGO_PKG_VERSION")))
+            .build()
+            .expect("a client with static options always builds")
+    })
+}
+
+/// Reads a body up to `cap` bytes; one byte more is an error, not a
+/// truncation, so a caller never parses half a document.
+pub async fn bytes(mut response: reqwest::Response, cap: usize) -> anyhow::Result<Vec<u8>> {
+    let hint = response.content_length().unwrap_or(0) as usize;
+    if hint > cap {
+        anyhow::bail!("response of {hint} bytes exceeds the cap of {cap}");
+    }
+    let mut out = Vec::with_capacity(hint.min(cap));
+    while let Some(chunk) = response.chunk().await? {
+        if out.len() + chunk.len() > cap {
+            anyhow::bail!("response exceeds the cap of {cap} bytes");
+        }
+        out.extend_from_slice(&chunk);
+    }
+    Ok(out)
+}
+
+pub async fn text(response: reqwest::Response, cap: usize) -> anyhow::Result<String> {
+    Ok(String::from_utf8(bytes(response, cap).await?)?)
+}
+
+/// `GET url`, status checked, body read under `cap`.
+pub async fn get_text(url: &str, cap: usize) -> anyhow::Result<String> {
+    let response = client().get(url).send().await?.error_for_status()?;
+    text(response, cap).await
+}

src/lib.rs

@@ -19,6 +19,9 @@ pub mod needs;
 #[cfg(feature = "ssr")]
 pub mod local;
 
+#[cfg(feature = "ssr")]
+pub mod http;
+
 #[cfg(feature = "ssr")]
 pub mod server;
 

src/main.rs

@@ -145,7 +145,13 @@ async fn main() -> anyhow::Result<()> {
         .route("/auth/callback", get(oidc::callback))
         .route("/auth/logout", get(oidc::logout))
         .route("/api/{*fn_name}", any(server_fn_handler))
-        .route("/upload", post(upload::upload))
+        // axum caps request bodies at 2 MB by default, which had made the
+        // 20 MB upload limit unreachable; raise the fence on this route only,
+        // with room for the multipart framing.
+        .route(
+            "/upload",
+            post(upload::upload).layer(axum::extract::DefaultBodyLimit::max(upload::MAX_UPLOAD_BYTES + 64 * 1024)),
+        )
         .route("/gitea-repo", get(content::gitea_repo_handler))
         // Content-shipped assets (hero module etc.), same-origin.
         .route("/site/{*path}", get(content::site_asset_handler))

src/resource.rs

@@ -260,21 +260,20 @@ async fn fetch_gitea_json(
     for (k, v) in params {
         url.query_pairs_mut().append_pair(k, v);
     }
-    let client = openidconnect::reqwest::Client::new();
-    let mut req = client.get(url.clone());
+    let mut req = crate::http::client().get(url.clone());
     if let Ok(token) = std::env::var("GITEA_API_TOKEN") {
         if !token.is_empty() {
             req = req.header("Authorization", format!("token {token}"));
         }
     }
-    let text = req
+    let response = req
         .send()
         .await
         .and_then(|r| r.error_for_status())
-        .map_err(|e| ServerFnError::new(format!("fetching {url}: {e}")))?
-        .text()
+        .map_err(|e| ServerFnError::new(format!("fetching {url}: {e}")))?;
+    let text = crate::http::text(response, crate::http::GITEA_CAP)
         .await
-        .map_err(|e| ServerFnError::new(e.to_string()))?;
+        .map_err(|e| ServerFnError::new(format!("reading {url}: {e}")))?;
     serde_json::from_str(&text).map_err(|e| ServerFnError::new(e.to_string()))
 }
 
@@ -341,16 +340,17 @@ async fn fetch_url_json(
     for (k, v) in &leftover {
         parsed.query_pairs_mut().append_pair(k, v);
     }
-    let client = openidconnect::reqwest::Client::new();
-    let text = client
+    let response = crate::http::client()
         .get(parsed.clone())
         .send()
         .await
         .and_then(|r| r.error_for_status())
-        .map_err(|e| ServerFnError::new(format!("fetching {parsed}: {e}")))?
-        .text()
+        .map_err(|e| ServerFnError::new(format!("fetching {parsed}: {e}")))?;
+    // Content picks the URL; the cap is what keeps a runaway endpoint
+    // from becoming a runaway allocation here.
+    let text = crate::http::text(response, crate::http::URL_RESOURCE_CAP)
         .await
-        .map_err(|e| ServerFnError::new(e.to_string()))?;
+        .map_err(|e| ServerFnError::new(format!("reading {parsed}: {e}")))?;
     serde_json::from_str(&text).map_err(|e| ServerFnError::new(e.to_string()))
 }
 

src/upload.rs

@@ -15,7 +15,7 @@ use serde::Serialize;
 
 use crate::server::AppState;
 
-const MAX_UPLOAD_BYTES: usize = 20 * 1024 * 1024;
+pub const MAX_UPLOAD_BYTES: usize = 20 * 1024 * 1024;
 
 #[derive(Clone)]
 pub struct Garage {
@@ -84,13 +84,21 @@ pub async fn upload(
             .unwrap_or("application/octet-stream")
             .to_string();
         let filename = field.file_name().unwrap_or("upload").to_string();
-        let bytes = field
-            .bytes()
+        // Chunk by chunk, refused the moment the cap is crossed: an
+        // oversized body is never buffered whole before being turned
+        // away. The route's body limit (main.rs) is the outer fence.
+        let mut bytes: Vec<u8> = Vec::new();
+        let mut field = field;
+        while let Some(chunk) = field
+            .chunk()
             .await
-            .map_err(|e| (StatusCode::BAD_REQUEST, e.to_string()))?;
-        if bytes.len() > MAX_UPLOAD_BYTES {
+            .map_err(|e| (StatusCode::BAD_REQUEST, e.to_string()))?
+        {
+            if bytes.len() + chunk.len() > MAX_UPLOAD_BYTES {
                 return Err((StatusCode::PAYLOAD_TOO_LARGE, "file too large".to_string()));
             }
+            bytes.extend_from_slice(&chunk);
+        }
 
         let key = format!("{}-{filename}", uuid::Uuid::new_v4());
         garage